1. Introduction
Caffeine [
1] (1,3,7-trimethyl-3,4,5,7-tetrahydro-1H-purine-2,6-dione) is one of the most popular neurostimulants in the world, and together with theophylline and theobromine, it belongs to methyl-xanthines. Its stimulating and hallucinogenic properties have been known to humanity since ancient times. From a chemical point of view, it is a purine alkaloid [
1], which is most often consumed as a component of tea and coffee [
2,
3]. Its molecular structure in the gas phase has been reported on [
1], and its chemical formula with atom labeling is shown in
Figure 1.
Recently, caffeine could be purchased in the form of concentrated solutions—so-called “energy drinks”—as well as tablets. This substance is routinely used in medicine as a hypervascular agent [
2]. In everyday life, caffeine is often consumed by schoolchildren, students and athletes to increase intellectual and physical performance. In addition, caffeine use plays a significant role in the work activities of some iconic representatives of science and art. Mathematician Paul Erdős attributed caffeine as a key factor in his working effectiveness, and after a 30-day pause in caffeine consumption, he had “trouble doing mathematics without coffee”, claiming that his “mathematical notes became blank pieces of paper” that he just “stared at, unable to work”. Another “connoisseur” of caffeine was the mathematician Alfred Renyi, the author of the aphorism “A mathematician is a machine for turning coffee into theorems”. It is also known that a famous painter W. H. Auden considered coffee to be one of the “labor-saving utensils” of his “intellectual kitchen”. Following his creative ambitions and struggling with frequent debts, the literary icon Honore de Balzac also drank large quantities (up to 50 cups per day) of strong coffee [
4,
5]. However, caffeine abuse does not lead to significant health problems, as in the case of cocaine or amphetamines or even heroin or morphine [
2]. Moreover, world and government health organizations have not registered cases of drug addiction caused by caffeine. For this reason, there are no restrictions on the production of caffeine-containing products and no exact standards for the content of this alkaloid in these products [
2]. However, some scientific sources demonstrate that caffeine abuse leads to negative, although reversible, consequences for human health. Concentration-dependent increases in blood pressure, urination frequency and anxiety were recorded [
2,
3]. In the case of people with diseases of the cardiovascular and urinary systems, these facts could not be neglected. Moreover, people with genetic predisposition and other factors that increase the risk of heart, vascular and kidney diseases should also pay special attention to the frequency and amount of caffeine consumption. For this reason, it is extremely important to develop methods that allow for the identification and quantification of the amount of caffeine contained in humans’ consumed products. The literature provides data on the use of various chromatographic and spectroscopic methods for this purpose—high-performance liquid chromatography (HPLC) with ultraviolet detection (UV) [
6,
7], mass spectrometry (MS) [
8,
9,
10], particle beam/electron ionization mass spectrometry (PB/EIMS) [
11], Fourier transform near-infrared reflectance (FT-NIR) spectroscopy [
12] and nuclear magnetic resonance (NMR) spectroscopy [
13,
14,
15].
As can be noticed, the history of experimental and theoretical studies on caffeine is quite diverse and extensive [
1,
3,
9,
10], but this stimulant is still open to newly designed chemical studies. One of the relatively new methods for the characterization of natural products’ properties is the in silico prediction of some of their parameters. Below, we will mention a few recent theoretical studies on the description of some structural, spectroscopic [
16] and chemical reactivity parameters on a free molecule of caffeine and in solution, using an approximate solvent effect only via a polarizable continuum model (PCM [
17,
18]). Gibson and Fowler [
19] studied the aromaticity of methyl-xanthines, including caffeine at low levels of theory (HF/6-31G* and B3LYP/6-31G*). They revealed a strong delocalized π-electron current above the imidazole ring of caffeine molecules. Another density functional theory (DFT [
20,
21]) study by Rijal et al. [
22] concentrated on nicotine and caffeine in the gas phase and solution employing B3LYP/6-311++G**. They discussed dipole moment changes in free molecules and the magnitude of changes as a result of solvent impacts. Additionally, they calculated global reactivity descriptors, density of states (DOSs), atomic charges and vibrational and electronic spectra. Gomez et al. [
23] reported on a DFT analysis of conformational preferences in caffeine, employing B3LYPD3/6-311++G** calculations with the single-point domain-based local pair natural orbital coupled cluster [
24] (DLPNO-CCSD(T)) refinement of energy at stationary points. They concluded that there was the presence of a nearly free rotation or fluctuation in the methyl groups in caffeine.
As can be seen from
Figure 1, caffeine contains hydrogen, carbon, nitrogen and oxygen elements in its structure. However, only experimental
1H,
13C and
15N NMR spectra [
25] have been reported. Thus, to our surprise, no
17O NMR spectra of caffeine are available in the literature [
25].
Several original experimental studies and review papers on the application of oxygen-17 NMR to characterize organic and inorganic molecules have been reported [
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36]. Additionally, theoretical predictions of
17O NMR have been also reported [
37,
38,
39]. Following Krivdin’s review, it is worth citing well-defined difficulties in running 17O NMR spectra: “The
17O NMR signals are difficult to detect even for small inorganic and organic molecules. Furthermore,
17O isotope has a very low natural abundance of only 0.038%, so that it is routinely observable only for neat liquids and solutions of very high concentrations or, alternatively, in the
17O-enriched samples.” Moreover, among the known deficiencies of
17O nuclei is the very low natural receptivity in comparison to protons (1.11 × 10
−5). Its spin of 5/2 leads to very broad lines, in particular in asymmetric environments. Low Larmor frequency results in acoustic ringing (rolling baseline). Therefore, highly concentrated solutions could be studied. Taking the above into account, the lack of natural abundance oxygen-17 NMR reports could be due to the difficulties of experimental studies of this isotope by using the NMR technique.
On the other hand, Colherinhas et al. [
40] recently reported on DFT calculated chemical shifts in caffeine in water using four density functionals (B3LYP, CAM-B3LYP, BHandHLYP and PBE1PBE) combined with a relatively small Pople-type 6-311++G(2d,2p) basis set. The authors reported on the B3LYP/aug-cc-pVDZ geometry optimization of caffeine in water using the PCM approach for the subsequent gauge-including atomic orbital (GIAO [
41,
42]) NMR calculation of nuclear magnetic shielding tensors. Interestingly, Colherinhas et al. [
40] used experimental data from Ulrich et al. [
43] for comparison with their predicted values. However, they did not report [
40] experimental chemical shifts in three different methyl carbons in caffeine molecules. Obviously, in our opinion, the authors [
40] could compare their theoretical chemical shifts to already-reported proton and carbon NMR data in chloroform [
25]. Additionally, in comparison to a more complete and flexible triple-zeta aug-cc-pVTZ Dunning-type basis set, they used [
40] a significantly smaller aug-cc-pVDZ one. This basis set is known to be very unreliable and prone to producing large errors [
44,
45].
For this reason, we decided to expand the existing knowledge about the spectroscopic properties of caffeine and present in this work its first experimental
17O NMR spectrum. In addition, the current development of analytical techniques makes it possible to discover more and more structurally similar plant metabolites. Quests for new substances, originating from both new and long-known plant species, are popular in the literature [
13]. In this case, molecular modeling methods are frequently used to support the determination of bioactive metabolite structure and interpret the spectroscopic properties of novel compounds. For this reason, in the current study, we are aiming at developing an efficient theoretical methodology for predicting reliable NMR properties of caffeine and similar alkaloids. Therefore, we tested nine density functionals, including B3LYP, BLYP, BP86, CAM-B3LYP, LC-BLYP, M06, PBE0, TPSSh and wB97X, and applied them to study the NMR properties of caffeine in vacuum and three solvents (chloroform, DMSO and water). To verify the accuracy of the calculated results, the theoretical predicted parameters were compared with reported
1H,
13C and
15N NMR data [
25], as well as with our experimental results for
17O NMR.
4. Conclusions
A combined experimental and theoretical study on the NMR spectroscopy of caffeine is reported on. For the first time, we report on the experimental 17O NMR spectra of caffeine in solution. The assignment of two oxygen-17 peaks of caffeine was supported by systematic DFT calculations. The solvent effect was included via an implicit PCM model, and both multinuclear isotropic magnetic shieldings (chemical shifts) and indirect spin–spin coupling constants were calculated in vacuum, chloroform, DMSO and water. Benchmark GIAO calculations using nine functionals (B3LYP, BLYP, BP86, CAM-B3LYP, LC-BLYP, M06, PBE0, TPSSh, wB97x) and a large aug-cc-pVTZ basis set predicted fairly reasonable chemical shifts for the 13C, 1H, 15N and 17O nuclei. The following smallest RMS deviations for caffeine nuclei in chloroform were calculated: 13C with TPSSh and BP86 of 3.14 and 3.84 ppm; 1H with B3LYP and PBE0 of 0.044 and 0.047 ppm; 15N with PBE0 and TPSSh of 3.14 and 5.22 ppm; and 17O with LC-BLYP and TPSSh of 15.62 and 9.66 ppm.
The prediction of 1JCH coupling was more efficient using CAM-B3LYP/aug-cc-pVTZ (mixed) calculations (RMS deviation of 0.66 and 2.31 Hz in vacuum and chloroform).
It is evident from the obtained data that no single density functional works best for all tested NMR parameters of purines. Therefore, an optimal methodology for calculating the chemical shifts and J-couplings of caffeine and similar molecules has been developed. We suggest using the TPSSh and BP86 density functionals to obtain reliable 13C NMR data, B3LYP and PBE0 for 1H, PBE0 and TPSSh for 15N, and LC-BLYP and TPSSh for 17O, suitable for studies of purines and similar natural products.